Super-high, scale invariant image compression using a surface learning neural network

This paper discusses a new method of using neural networks for super-high, scale invariant image compression. This is achieved by training an multilayer perceptron (MLP) network to learn an approximation to a two-dimensional image surface. One of the main features of this network is its ability to perform unsupervised feature extraction and in so doing represent image features in a compact form the results demonstrate the potential for significantly superior edge reproduction at super-high compression as compared to standard subband coding methods